Shared SNP effects across breeds increase the genomic prediction accuracy for numerically small breeds.

Genomic selection has proven effective for accelerating genetic improvement in livestock, but its application in developing countries faces challenges, particularly for numerically small breeds, wherein, establishing large, single-breed reference population is difficult. To address this limitation, current study aimed to investigate the effectiveness of multi-breed reference populations to improve the genomic prediction accuracy in numerically small breeds with limited phenotypic and genomic resources. Genotypic and phenotypic data from 1,298 Gir, 1,291 Sahiwal, and 500 Kankrej indigenous cattle were used to enhance the genomic prediction accuracy in Kankrej, utilizing a multi-breed reference population. Principal Component Analysis (PCA)-K-means-based clustering showed overlap between Gir and Kankrej, indicating genetic similarity between these two breeds. Linkage disequilibrium (LD) decay patterns corroborated these results, showing a similar trend in the LD decay plot in the Gir-Kankrej, suggesting shared haplotype blocks that can be utilized in combined analyses. Heritability estimates for 305-day first lactation milk yield (305-DMY) were 0.30 ± 0.07 for Gir, 0.27 ± 0.07 for Sahiwal, and 0.17 ± 0.01 for Kankrej cattle. Genomic estimated breeding values (GEBV) with single-breed reference population were predicted with accuracies of 0.65 for Gir, 0.60 for Sahiwal, and 0.49 for Kankrej. Multi-breed reference populations were created for all possible breed combinations. Three genomic evaluation strategies, viz., shared Genomic Relationship Matrix (GRM), non-shared GRM, and metafounder-corrected shared GRM, were compared. We observed a significant gain in accuracy for the numerically small breed, Kankrej, using a multi-breed approach as compared to a single-breed approach. Using a multi-breed evaluation with shared and non-shared GRM and the metafounder approach led to accuracy improvements of 23.6%, 24.6%, and 16.9%, respectively, while using the Gir-Kankrej multi-breed reference population. Validation using the linear regression (LR) method showed that the best way to predict GEBV for the Kankrej breed was through multi-breed evaluation with the 'shared GRM' approach. Our findings indicate that in the absence of large breed specific reference, multibreed genomic evaluation offers a viable strategy for enhancing genomic prediction accuracy for numerically small breeds in India.